SPSS Statistics

Hey all, I am writing my finance bachelor's thesis on the impact of Covid-19 on household's portfolio choices across different wealth groups in the BeNeLux area (Belgium, Netherlands, Luxembourg). The data comes from European Central Bank: (https://www.ecb.europa.eu/stats/ecb_surveys/hfcs/html/index.en.html) and consists of different financial figures of households separated per country and per household wealth group (6 groups: bottom 20%, 20-40%, 40-60%, 60-80%, 80-90%, 90-100%). I have data of 4 waves (2011, 2014, 2017, 2021). With these years being the independent variable (2021 as year of focus since this is mid-covid). Besides just plotting the figures in graphs to check for any significant changes, I would like to run some statistical tests and regressions to test the significance of any differences of the year 2021 to the other three waves (2011, 2014, 2017)

Figures I will mainly focus on include:
A3 Net wealth, medians
A4 Net wealth, means
B3 Real assets, ownership of HMR
B5 Real estate assets, conditional medians
C4 Financial assets, conditional medians
C5 Financial assets, has shares
D4 Share of financial assets on total assets
E5 Percentage of households holding debt
E6 Total debt, conditional medians
F4 Median debt to income ratio
F5 Median debt service to income ratio, among households with debt payments
F6 Median debt to assets ratio – breakdowns
G3 Regular expenses less than income

As you can see these figures consist of medians, means, ratio's and %'s of each seperate wealth group. I do have the standard errors for each datapoint (so for each country each separate group figure)

With these figures being aggregate data from a large survey, I am not sure which statistical tests and what kind of regressions I can use. I heard from my supervisor to aim for 30-50 datapoints per regression, however my data only consists of figures (means, medians, ratio's) of 6 large groups. This would leave me with 6 data points per country per financial figure, so 18 datapoints per financial figure per year, so 72 datapoints per financial figure across the 4 years. With these figures being aggregate data, do these datapoints suffice for a regression analysis? (if so, which type?)

Could anyone advise me on which statistical tests and regressions to use with this data, to check whether the year 2021 is significantly different from the others, other than just plotting graphs? Thanks in advance.

My first question is whether you can get at the raw data from the surveys. If not, consider using a decision tree to parse out the relative levels of each of your measurements to see if you get a clearer picture of the data. What is your dependent variable?

Hey all, I am writing my finance bachelor's thesis on the impact of Covid-19 on household's portfolio choices across different wealth groups in the BeNeLux area (Belgium, Netherlands, Luxembourg). The data comes from European Central Bank: (https://www.ecb.europa.eu/stats/ecb_surveys/hfcs/html/index.en.html) and consists of different financial figures of households separated per country and per household wealth group (6 groups: bottom 20%, 20-40%, 40-60%, 60-80%, 80-90%, 90-100%). I have data of 4 waves (2011, 2014, 2017, 2021). With these years being the independent variable (2021 as year of focus since this is mid-covid). Besides just plotting the figures in graphs to check for any significant changes, I would like to run some statistical tests and regressions to test the significance of any differences of the year 2021 to the other three waves (2011, 2014, 2017)

Figures I will mainly focus on include:
A3 Net wealth, medians
A4 Net wealth, means
B3 Real assets, ownership of HMR
B5 Real estate assets, conditional medians
C4 Financial assets, conditional medians
C5 Financial assets, has shares
D4 Share of financial assets on total assets
E5 Percentage of households holding debt
E6 Total debt, conditional medians
F4 Median debt to income ratio
F5 Median debt service to income ratio, among households with debt payments
F6 Median debt to assets ratio – breakdowns
G3 Regular expenses less than income

As you can see these figures consist of medians, means, ratio's and %'s of each seperate wealth group. I do have the standard errors for each datapoint (so for each country each separate group figure)

With these figures being aggregate data from a large survey, I am not sure which statistical tests and what kind of regressions I can use. I heard from my supervisor to aim for 30-50 datapoints per regression, however my data only consists of figures (means, medians, ratio's) of 6 large groups. This would leave me with 6 data points per country per financial figure, so 18 datapoints per financial figure per year, so 72 datapoints per financial figure across the 4 years. With these figures being aggregate data, do these datapoints suffice for a regression analysis? (if so, which type?)

Could anyone advise me on which statistical tests and regressions to use with this data, to check whether the year 2021 is significantly different from the others, other than just plotting graphs? Thanks in advance.

Dear Marianne,

Raw level data of household finances is quite confidential, and me as a bachelor's thesis writer I do not think I will be able to get access. Besides that, requesting raw level data now might be a bit too late and is maybe not the best idea due to time constraints of handing in my thesis.

How can I use a decision tree to parse out the relative levels of my financial figures? I believe it is some kind of machine learning technique? Does this require pre-knowledge of programming for example?

My aim is to examine the impact of the covid-19 pandemic on household portfolio choice, however this doesn't consist of one exact measure, so I have multiple dependent variables I want to test which influence household portfolio choice, these mostly consist of the before mentioned financial figures, ratio's and percentages such as: share of financial assets on total assets, percentage of households holding debt, real estate assets, etc.

I want to emphasize that I do have the standard errors for each datapoint (so for each country each separate group figure)

My first question is whether you can get at the raw data from the surveys. If not, consider using a decision tree to parse out the relative levels of each of your measurements to see if you get a clearer picture of the data. What is your dependent variable?

Hey all, I am writing my finance bachelor's thesis on the impact of Covid-19 on household's portfolio choices across different wealth groups in the BeNeLux area (Belgium, Netherlands, Luxembourg). The data comes from European Central Bank: (https://www.ecb.europa.eu/stats/ecb_surveys/hfcs/html/index.en.html) and consists of different financial figures of households separated per country and per household wealth group (6 groups: bottom 20%, 20-40%, 40-60%, 60-80%, 80-90%, 90-100%). I have data of 4 waves (2011, 2014, 2017, 2021). With these years being the independent variable (2021 as year of focus since this is mid-covid). Besides just plotting the figures in graphs to check for any significant changes, I would like to run some statistical tests and regressions to test the significance of any differences of the year 2021 to the other three waves (2011, 2014, 2017)

Figures I will mainly focus on include:
A3 Net wealth, medians
A4 Net wealth, means
B3 Real assets, ownership of HMR
B5 Real estate assets, conditional medians
C4 Financial assets, conditional medians
C5 Financial assets, has shares
D4 Share of financial assets on total assets
E5 Percentage of households holding debt
E6 Total debt, conditional medians
F4 Median debt to income ratio
F5 Median debt service to income ratio, among households with debt payments
F6 Median debt to assets ratio – breakdowns
G3 Regular expenses less than income

As you can see these figures consist of medians, means, ratio's and %'s of each seperate wealth group. I do have the standard errors for each datapoint (so for each country each separate group figure)

With these figures being aggregate data from a large survey, I am not sure which statistical tests and what kind of regressions I can use. I heard from my supervisor to aim for 30-50 datapoints per regression, however my data only consists of figures (means, medians, ratio's) of 6 large groups. This would leave me with 6 data points per country per financial figure, so 18 datapoints per financial figure per year, so 72 datapoints per financial figure across the 4 years. With these figures being aggregate data, do these datapoints suffice for a regression analysis? (if so, which type?)

Could anyone advise me on which statistical tests and regressions to use with this data, to check whether the year 2021 is significantly different from the others, other than just plotting graphs? Thanks in advance.

Hi, Lukas, When you say "impact", do you mean that as a ranked variable, a binary variable, or a scalar variable? Once you define that, I can help.

Dear Marianne,

Raw level data of household finances is quite confidential, and me as a bachelor's thesis writer I do not think I will be able to get access. Besides that, requesting raw level data now might be a bit too late and is maybe not the best idea due to time constraints of handing in my thesis.

How can I use a decision tree to parse out the relative levels of my financial figures? I believe it is some kind of machine learning technique? Does this require pre-knowledge of programming for example?

My aim is to examine the impact of the covid-19 pandemic on household portfolio choice, however this doesn't consist of one exact measure, so I have multiple dependent variables I want to test which influence household portfolio choice, these mostly consist of the before mentioned financial figures, ratio's and percentages such as: share of financial assets on total assets, percentage of households holding debt, real estate assets, etc.

I want to emphasize that I do have the standard errors for each datapoint (so for each country each separate group figure)

My first question is whether you can get at the raw data from the surveys. If not, consider using a decision tree to parse out the relative levels of each of your measurements to see if you get a clearer picture of the data. What is your dependent variable?

Hey all, I am writing my finance bachelor's thesis on the impact of Covid-19 on household's portfolio choices across different wealth groups in the BeNeLux area (Belgium, Netherlands, Luxembourg). The data comes from European Central Bank: (https://www.ecb.europa.eu/stats/ecb_surveys/hfcs/html/index.en.html) and consists of different financial figures of households separated per country and per household wealth group (6 groups: bottom 20%, 20-40%, 40-60%, 60-80%, 80-90%, 90-100%). I have data of 4 waves (2011, 2014, 2017, 2021). With these years being the independent variable (2021 as year of focus since this is mid-covid). Besides just plotting the figures in graphs to check for any significant changes, I would like to run some statistical tests and regressions to test the significance of any differences of the year 2021 to the other three waves (2011, 2014, 2017)

Figures I will mainly focus on include:
A3 Net wealth, medians
A4 Net wealth, means
B3 Real assets, ownership of HMR
B5 Real estate assets, conditional medians
C4 Financial assets, conditional medians
C5 Financial assets, has shares
D4 Share of financial assets on total assets
E5 Percentage of households holding debt
E6 Total debt, conditional medians
F4 Median debt to income ratio
F5 Median debt service to income ratio, among households with debt payments
F6 Median debt to assets ratio – breakdowns
G3 Regular expenses less than income

As you can see these figures consist of medians, means, ratio's and %'s of each seperate wealth group. I do have the standard errors for each datapoint (so for each country each separate group figure)

With these figures being aggregate data from a large survey, I am not sure which statistical tests and what kind of regressions I can use. I heard from my supervisor to aim for 30-50 datapoints per regression, however my data only consists of figures (means, medians, ratio's) of 6 large groups. This would leave me with 6 data points per country per financial figure, so 18 datapoints per financial figure per year, so 72 datapoints per financial figure across the 4 years. With these figures being aggregate data, do these datapoints suffice for a regression analysis? (if so, which type?)

Could anyone advise me on which statistical tests and regressions to use with this data, to check whether the year 2021 is significantly different from the others, other than just plotting graphs? Thanks in advance.

Maybe a combination of more types of variables, since the 'impact' can not be identified as one variable?

I am thinking of comparing numerous figures to ideally discover a significant change in the covid year compared to its previous years.
These variables can consist of:
1. Ranked variables, which I would need to define myself (e.g. level of risk aversion based on their portfolio composition)
2. Categorical variables (e.g. decrease investment in stocks, increased preference for real estate, etc.)
3. Scalar variables (e.g. significant percentage changes in financial figures, ratio's)

I haven't decided how to measure 'impact' exactly, but variables above should give you a better idea of what I can be looking for.
Ultimately I want to discover various financial figures which were stable before covid-19, that had a significant change during covid-19, to increase the probability of the change being the result of covid-19.

Others have pointed me to use these statistical tests; Hotelling's t-square test, weighted least squares, Fisher's Linear Discriminant Analysis

Hi, Lukas, When you say "impact", do you mean that as a ranked variable, a binary variable, or a scalar variable? Once you define that, I can help.

Dear Marianne,

Raw level data of household finances is quite confidential, and me as a bachelor's thesis writer I do not think I will be able to get access. Besides that, requesting raw level data now might be a bit too late and is maybe not the best idea due to time constraints of handing in my thesis.

How can I use a decision tree to parse out the relative levels of my financial figures? I believe it is some kind of machine learning technique? Does this require pre-knowledge of programming for example?

My aim is to examine the impact of the covid-19 pandemic on household portfolio choice, however this doesn't consist of one exact measure, so I have multiple dependent variables I want to test which influence household portfolio choice, these mostly consist of the before mentioned financial figures, ratio's and percentages such as: share of financial assets on total assets, percentage of households holding debt, real estate assets, etc.

I want to emphasize that I do have the standard errors for each datapoint (so for each country each separate group figure)

My first question is whether you can get at the raw data from the surveys. If not, consider using a decision tree to parse out the relative levels of each of your measurements to see if you get a clearer picture of the data. What is your dependent variable?

Hey all, I am writing my finance bachelor's thesis on the impact of Covid-19 on household's portfolio choices across different wealth groups in the BeNeLux area (Belgium, Netherlands, Luxembourg). The data comes from European Central Bank: (https://www.ecb.europa.eu/stats/ecb_surveys/hfcs/html/index.en.html) and consists of different financial figures of households separated per country and per household wealth group (6 groups: bottom 20%, 20-40%, 40-60%, 60-80%, 80-90%, 90-100%). I have data of 4 waves (2011, 2014, 2017, 2021). With these years being the independent variable (2021 as year of focus since this is mid-covid). Besides just plotting the figures in graphs to check for any significant changes, I would like to run some statistical tests and regressions to test the significance of any differences of the year 2021 to the other three waves (2011, 2014, 2017)

Figures I will mainly focus on include:
A3 Net wealth, medians
A4 Net wealth, means
B3 Real assets, ownership of HMR
B5 Real estate assets, conditional medians
C4 Financial assets, conditional medians
C5 Financial assets, has shares
D4 Share of financial assets on total assets
E5 Percentage of households holding debt
E6 Total debt, conditional medians
F4 Median debt to income ratio
F5 Median debt service to income ratio, among households with debt payments
F6 Median debt to assets ratio – breakdowns
G3 Regular expenses less than income

As you can see these figures consist of medians, means, ratio's and %'s of each seperate wealth group. I do have the standard errors for each datapoint (so for each country each separate group figure)

With these figures being aggregate data from a large survey, I am not sure which statistical tests and what kind of regressions I can use. I heard from my supervisor to aim for 30-50 datapoints per regression, however my data only consists of figures (means, medians, ratio's) of 6 large groups. This would leave me with 6 data points per country per financial figure, so 18 datapoints per financial figure per year, so 72 datapoints per financial figure across the 4 years. With these figures being aggregate data, do these datapoints suffice for a regression analysis? (if so, which type?)

Could anyone advise me on which statistical tests and regressions to use with this data, to check whether the year 2021 is significantly different from the others, other than just plotting graphs? Thanks in advance.

It's hard to start testing data unless you know what your dependent variable is. Since you are still exploring, start with cluster analysis to see if there is a pattern that leads you to obvious dependent variables. For instance, you may find that stock market investment turns out to be the clearest delineator among your segments.

After that, discriminant analysis might help but you'll need to know what it's doing and what the results mean. something simpler, like a decision tree, might take less time to understand. After all, your goal is to get a degree, not do this all the time. But with both of these, you need a defined dependent variable.

If you want to use all of your dependent variables at once, see here: SPSS #34 - Multivariate Anova - 1 Independent & 2 Dependent (MANOVA)

YouTube		remove preview
SPSS #34 - Multivariate Anova - 1 Independent & 2 Dependent (MANOVA) View this on YouTube >

 You are reminding me of why I make some money helping PhD students complete their theses! 

Maybe a combination of more types of variables, since the 'impact' can not be identified as one variable?

I am thinking of comparing numerous figures to ideally discover a significant change in the covid year compared to its previous years.
These variables can consist of:
1. Ranked variables, which I would need to define myself (e.g. level of risk aversion based on their portfolio composition)
2. Categorical variables (e.g. decrease investment in stocks, increased preference for real estate, etc.)
3. Scalar variables (e.g. significant percentage changes in financial figures, ratio's)

I haven't decided how to measure 'impact' exactly, but variables above should give you a better idea of what I can be looking for.
Ultimately I want to discover various financial figures which were stable before covid-19, that had a significant change during covid-19, to increase the probability of the change being the result of covid-19.

Others have pointed me to use these statistical tests; Hotelling's t-square test, weighted least squares, Fisher's Linear Discriminant Analysis

Hi, Lukas, When you say "impact", do you mean that as a ranked variable, a binary variable, or a scalar variable? Once you define that, I can help.

Dear Marianne,

Raw level data of household finances is quite confidential, and me as a bachelor's thesis writer I do not think I will be able to get access. Besides that, requesting raw level data now might be a bit too late and is maybe not the best idea due to time constraints of handing in my thesis.

How can I use a decision tree to parse out the relative levels of my financial figures? I believe it is some kind of machine learning technique? Does this require pre-knowledge of programming for example?

My aim is to examine the impact of the covid-19 pandemic on household portfolio choice, however this doesn't consist of one exact measure, so I have multiple dependent variables I want to test which influence household portfolio choice, these mostly consist of the before mentioned financial figures, ratio's and percentages such as: share of financial assets on total assets, percentage of households holding debt, real estate assets, etc.

I want to emphasize that I do have the standard errors for each datapoint (so for each country each separate group figure)

My first question is whether you can get at the raw data from the surveys. If not, consider using a decision tree to parse out the relative levels of each of your measurements to see if you get a clearer picture of the data. What is your dependent variable?

Hey all, I am writing my finance bachelor's thesis on the impact of Covid-19 on household's portfolio choices across different wealth groups in the BeNeLux area (Belgium, Netherlands, Luxembourg). The data comes from European Central Bank: (https://www.ecb.europa.eu/stats/ecb_surveys/hfcs/html/index.en.html) and consists of different financial figures of households separated per country and per household wealth group (6 groups: bottom 20%, 20-40%, 40-60%, 60-80%, 80-90%, 90-100%). I have data of 4 waves (2011, 2014, 2017, 2021). With these years being the independent variable (2021 as year of focus since this is mid-covid). Besides just plotting the figures in graphs to check for any significant changes, I would like to run some statistical tests and regressions to test the significance of any differences of the year 2021 to the other three waves (2011, 2014, 2017)

Figures I will mainly focus on include:
A3 Net wealth, medians
A4 Net wealth, means
B3 Real assets, ownership of HMR
B5 Real estate assets, conditional medians
C4 Financial assets, conditional medians
C5 Financial assets, has shares
D4 Share of financial assets on total assets
E5 Percentage of households holding debt
E6 Total debt, conditional medians
F4 Median debt to income ratio
F5 Median debt service to income ratio, among households with debt payments
F6 Median debt to assets ratio – breakdowns
G3 Regular expenses less than income

As you can see these figures consist of medians, means, ratio's and %'s of each seperate wealth group. I do have the standard errors for each datapoint (so for each country each separate group figure)

With these figures being aggregate data from a large survey, I am not sure which statistical tests and what kind of regressions I can use. I heard from my supervisor to aim for 30-50 datapoints per regression, however my data only consists of figures (means, medians, ratio's) of 6 large groups. This would leave me with 6 data points per country per financial figure, so 18 datapoints per financial figure per year, so 72 datapoints per financial figure across the 4 years. With these figures being aggregate data, do these datapoints suffice for a regression analysis? (if so, which type?)

Could anyone advise me on which statistical tests and regressions to use with this data, to check whether the year 2021 is significantly different from the others, other than just plotting graphs? Thanks in advance.